I still remember Intel’s Developer Forum in 2008, before the company had launched its Core i7 ‘Bloomfield’ lineup on LGA 1366. All of the demo machines Intel was showing off behind closed doors were driven by its then-unreleased SSDs. And I still remember Francois Piednoel noting that, without SSDs, the performance of Core i7 would be handicapped by conventional hard drives. Since then, we’ve tied solid state storage into as much of our high-end processor and graphics testing as possible.
As you no doubt already know, hard drives and SSDs alike have almost universally connected to the rest of your system via 3 Gb/s SATA ports. The standard has been fairly accessible since 2005, so most remotely-modern machines already support it.
Three gigabits per second divided by eight gives you a maximum throughput of 375 megabytes per second. However, 8b/10b encoding exacts an immediate 20% bandwidth penalty, dropping the ceiling of a single SATA 3 Gb/s port to 300 MB/s. That’s actually a fairly large pipe. It’s large enough, in fact, that no single mechanical hard drive is able to saturate it (Patrick has an upcoming review of Western Digital’s 2TB Caviar Black where it pushes about 140 MB/s in sequential reads). It takes a current-generation SSD performing sequential reads to hit the limits of what SATA 3 Gb/s can do. And even then, we’d hardly call the latest drives bottlenecked by their SATA interface.
Yet, here I sit, staring at an Asus motherboard with Marvell’s 88SE9128 6 Gb/s SATA controller, Seagate’s behemoth 2TB Barracuda XT 6 Gb/s SATA drive, and an engineering sample of Marvell’s upcoming 6 Gb/s SSD controller.
And when I say engineering sample, I mean very, very early hardware. As in, this SSD sports a single-digit serial number and doesn’t include any NAND flash memory. Thus, we’re only able to test the controller itself using a limited number of benchmarks. I can't even include a picture of the thing's guts without raising hell.
I don’t have time to do a formal review of the Asus motherboard or the Seagate drive (my flight to Spain leaves in 10 hours). Thomas will be taking care of Asus' P55-based P7P55D Premium motherboard soon, and Patrick plans to cover the Barracuda XT in an upcoming feature. Today, I’m able to offer a quick look at how these three components work together and a brief peek at performance.
Update: It's worth noting that, with 600 MB/s possible per SATA 6 Gb/s port, the P55's second-gen PCI Express links (limited to 2.5 GT/s each) are really insufficient. Asus gets around this on the P7P55D Premium with a PLX8613 second-gen bridge chip, which takes four of the PCH's links and turns them into a single 500 MB/s connection to Marvell's SATA controller. In theory, there's still a potential bottleneck there. But as we'll see, even today's storage devices are really quite unable to saturate it.
A First Look At SATA 6 Gb/s
Before we start playing with Marvell’s latest controller, let’s have a look at the first 6 Gb/s SATA hard drive to hit our labs: Seagate’s Barracuda XT. The 2TB monster spins at 7,200 RPM and sports a 64MB data buffer, similar to Western Digital’s Caviar Black (though the 2TB Caviar is limited to 3 Gb/s signaling).
Because it supports SATA 6 Gb/s transfer rates, Seagate is able to claim a maximum interface speed of up to 600 MB/s—twice what SATA 3 Gb/s could do. Of course, it’s wholly unrealistic to expect the drive to come anywhere close. It boasts the same areal density and similar raw data rates as the Barracuda 7200.12s we benchmarked earlier this year; performance should be, for the most part, fairly similar.
In order to zero in on any gain enabled by the 6 Gb/s interface, we tested the Barracuda XT on Marvell’s 88SE9128 and then again on Intel’s P55 PCH, with write-cache buffer flushing enabled and disabled in both environments.
|Header Cell - Column 0||Write-Cache Buffer Flushing Enabled||Write-Cache Buffer Flushing Disabled|
|6 Gb/s Marvell Controller||5495||6420|
|3 Gb/s Intel P55 Controller||5307||6045|
In PCMark Vantage, two things are apparent. First, there’s a significant performance improvement to be had when you disable Windows write-cache buffer flushing (it’s enabled by default), and second, regardless of whether you have the aforementioned feature turned on or off, the 6 Gb/s interface helps Seagate’s Barracuda XT out a bit.
|Header Cell - Column 0||6 Gb/s Marvell Controller||3 Gb/s Intel Controller|
|Sequential Read Performance||138.5 MB/s||138.5 MB/s|
|Random Access Speed||17.12ms||17.13 ms|
We get the same sequential read throughput and read access results in H2benchw.
|Everest Ultimate Edition|
|Row 0 - Cell 0||6 Gb/s Marvell Controller||3 Gb/s Intel Controller|
|Linear Read (Beginning)||140.1 MB/s||140.1 MB/s|
|Linear Read (Middle)||113.8 MB/s||114.5 MB/s|
|Linear Read (End)||64.4 MB/s||64.4 MB/s|
|Random Read||101.7 MB/s||73.3 MB/s|
|Read Access||16.64 ms||17.05 ms|
My next mobo MUST have:
SATA 3.0 6GB/s and USB3.0... PCIe 3.0 would be a bonus.
Hah, I wish! Just fortunate to be able to work with some cool people and cool hardware. First real vacation since my honeymoon almost five years ago!
Cheers from Barcelona!